Artificial Intelligence Driven Material Design for Porous Materials

Abstract

AbstractIn general, material properties and the underlaying microstructure are linked to each other. It is a frontier challenge to understand the associated structure-property relationship, which displays an essential ingredient for accelerated material design. Herein, we approach this issue with a unique machine learning assisted material design workflow, suitable to tailor the electrical conductivity based on the 3D microstructure or vice versa, in porous copper. Specifically, we integrate a multi-variable linear regression model for the targeted prediction and utilize a U-Net deep learning architecture to accurately classify the collected 3D image data. The evaluated 3D microstructure features and the electrical conductivity are used as an input for the prediction model. We show that the prediction reaches a maximum r-squared value of about 0.94. Our results highlight the importance of accurately retrieving a set of physical scrutinized microstructure features with statistical confidence, a key to conclude about the microstructure-property relationship.